Is the path to the economic transformation of intelligent agents Web3 intelligence or AI decentralization? Coinbase's launch of the Agentic Wallet may provide key answers. With the continuous advancement of AI agent frameworks over the past year, the application of agents is no longer limited to being a part of workflows involving human participation. Previously, our requirement for agents was to connect models to tools and processes, helping users quickly complete complex tasks under human oversight, but ultimately, the quality of the task output and the final decision-making power were controlled by the user. Recently, projects like OpenClaw and its derivative Moltbook indicate that agents have entered the next stage, where multiple agents autonomously divide tasks and actively cooperate with minimal supervision (or even no human supervision). This represents a significant enhancement in the capabilities and complexity of agents. However, they generally face a bottleneck: they are not entirely trustworthy when executing economically valuable tasks or tasks involving real assets. When tasks involve value creation or the flow of funds, they must pause to wait for human approval, which slows down the rapid evolution of the agent network. Against this backdrop, Coinbase has launched the Agentic Wallet, a wallet facility designed for agents, aiming to build a secure, controllable, and programmable smart economic infrastructure for autonomous agents. In fact, Coinbase had already launched another wallet tool for agents, AgentKit, last year. Its core design philosophy is to integrate the wallet into the agent's code, granting agents the ability to interact with the on-chain environment, allowing them to have a clear on-chain identity to participate in contract deployment, protocol operations, and other tasks. However, the primary goal of AgentKit's design is to introduce agents into the on-chain environment and bind them with a recognizable identity to participate in on-chain activities alongside humans. This design has certain flaws. First, the activity scope of agents is limited (restricted to on-chain), and each addition of an external system requires developers to perform a new integration, making it difficult to quickly expand application scenarios. Second, the responsibility for wallet private key management falls directly on developers and the agent operating environment. In cases of tool abuse or prompt injection attacks, the wallet permissions of agents can become a vulnerable attack surface. Therefore, AgentKit is more suitable for granting agents permission for on-chain interactions in low-complexity AI workflows involving human participation, but it faces collaboration challenges in networks composed of multiple agents. The core demand of the newly released Agentic Wallet is no longer to deeply participate in on-chain activities but to provide simple and secure access to on-chain assets and payment capabilities. Its design philosophy has shifted to "an independent wallet that any agent can call." It is no longer an SDK that requires deep integration but an independent wallet service that can be called through command-line tools or MCP services. The wallet can hold USDC and make payment requests through the x402 protocol, but agents will not have access to the wallet's private keys and do not need to handle complex transaction signing logic. All wallet security risks are abstracted and encapsulated within the backend services provided by Coinbase. This shift not only indicates that the large, all-encompassing single agent in the agent ecosystem is being replaced by a collaborative network composed of multiple specialized agents, but also suggests that individual agents cannot be fully trusted when managing real-world assets. Collaboration with other mature Web3 technologies is needed to control assets (such as backend TEE and signing, etc.). ✜ The preview section has ended; the remaining hidden core content is here👇